Model Card for pqc_guard

Model Details

• Model ID: rafayishaCked/pqc_gaurd
• Model Description: A fine-tuned Phi-2 model specialized in Post-Quantum Cryptography (PQC) reasoning and knowledge tasks.
• Developed by: Abdul Rafay
• Shared by: Abdul Rafay and M Masoom Alam
• Model type: phi-2 fine-tuned with LoRA adapters for PQC domain knowledge
• Language(s) (NLP): English
• Finetuned from model: microsoft/phi-2

---

Model Sources

• Repository (Hugging Face): https://huggingface.co/rafayishaCked/pqc_fine_tuned_model
• Training Notebook (Kaggle): Kaggle Training Code & Output
• Evaluation Notebook (Kaggle)L: [https://www.kaggle.com/code/abdulrafay07/evaluation]

---

Uses

Direct Use

• Intended for educational and research purposes in post-quantum cryptography.
• Can be used to answer PQC-related questions, simulate PQC scenarios, and support cryptography coursework or CTF-style tasks.

Out-of-Scope Uses

• Not intended for production cryptographic implementations.
• Not a replacement for official PQC standards or libraries.

---

Training Details

• Base model: microsoft/phi-2
• Fine-tuning approach: LoRA adapter fine-tuning
• Data: PQC-focused dataset covering lattice-based crypto, Kyber, Dilithium, etc.
• Compute: Trained on Kaggle (T4 GPU x2)
• Checkpoints: Uploaded at checkpoint-1600 and checkpoint-2316

---

Limitations

• Knowledge limited to training dataset scope.
• Overall a good model but restricted to PQC Domain.
• May hallucinate or misinterpret PQC standards.

---

Citation

@misc{alam2025pqcguard, author = {Muhammad Masoom Alam and Abdul Rafay}, title = {PQC-Guard: Fine-Tuning Phi-2 for Post-Quantum Cryptography Readiness}, year = {2025}, url = {https://github.com/abdulrafayishaCkER/pqc-guard} }